|
From my understanding of the genetic algorithm the population consists of individuals, where each individual is a potential solution made up of "genes", and each gene is a variable. So for a cost function that takes in 3 variables the individuals will have 3 genes. The optimization happens by selection (picking the fittests from the population), crossover (swapping genes between pairs in the population), and mutation (randomly changing certain genes). Because of this it seems to me that the genetic algorithm is best suited to high dimensionality optimisation problems, i.e ones with lots of variables, so that there are more genes that can be swapped and mutated. Is this true? and what is a good number of genes for use in the genetic algorithm? Thanks! |
|
Genetic Algorithms is a fancy way to do random search. There is little if no challenge in optimizing things with a small number of variables using genetic algorithms, in the 80s it made sense, since we did not have that much computational power, but nowadays you can just bruteforce the algorithm to reach global optimums. Genetic Algorithms are mostly used when you have little information of your system and wish to optimize its variables, like you know a cots function that you are interested in for which the mathematical derivation is hard, so you essentially just check which combination of parameters is better. As for the size of the gene chain, I do not think is better suited to any of them, if anything it just makes it harder to achieve to sensible results as you have more variables, since your mutation capabilities are undermined by your large number of variables (the mutation of a single gene, might have a negligible effect on the overall optimization), also, measure theory tells us that the distance between high dimensional variables looses information capabilities. |
